Why Ai Chatbots Make Bad Teachers - And How Teachers Can Exploit That Weakness

SEAN GLADWELL/Getty Images

ZDNET's cardinal takeaways

• ChatGPT Study Mode offers small to separate it from normal ChatGPT.
• A Study Mode personification must activity difficult to make nan instruction absorbing and rewarding.
• Developers and educators should push AI to stimulate students' curiosity.

In nan astir 3 years since OpenAI's ChatGPT burst onto nan scene, nan usage of artificial intelligence has invaded not only regular activity and leisure but besides nan section of education. 

The Pew Research Center reports that a 4th of US adults consult a bot to study something, up from 8% successful 2023 -- almost arsenic galore arsenic usage them for work. The percent is higher nan higher one's level of acquisition is, interestingly. 

That surge successful usage had put teachers and professors successful a quandary astir students' propensity to usage bots to get answers to questions alternatively than deliberation done problems deeply. Pew reports that a ample number of teachers fearfulness a situation successful education. 

Also: ChatGPT's study mode could beryllium your adjacent tutor - and it's free

Somewhat much nuanced, nan scholarly diary Daedalus, successful its rumor connected trends successful education, past twelvemonth concluded that "We cannot foretell really acquisition will beryllium affected successful nan agelong word by ample connection models and different AI-supported tools, but they clasp nan anticipation to some beforehand and distort existent approaches to school and learning."

So, what are educators -- and AI developers -- to do if nan world has recovered a measurement astir accepted education?

OpenAI's reply to nan mobility is simply a caller characteristic introduced to ChatGPT past week, Study Mode, which my workfellow Sabrina Ortiz explored. As Sabrina relates, Study Mode will respond to a punctual pinch a scheme of study and inquire questions astir goals. 

(Disclosure: Ziff Davis, ZDNET's genitor company, revenge an April 2025 suit against OpenAI, alleging it infringed Ziff Davis copyrights successful training and operating its AI systems.)

Also: How ChatGPT really useful (and why it's been truthful game-changing)

How I utilized Study Mode to effort to study a language

I tested Study Mode arsenic a measurement to study a caller language. I chose that activity because I've already utilized ChatGPT for a twelvemonth now to effort to study languages, which gave maine a ground for comparison. 

In my experience, Study Mode adds very small to my efforts to study a language. The differences pinch plain aged ChatGPT are minor. And I had to make galore efforts to steer Study Mode successful nan correct direction. 

The instruction present is clear. As pinch each connection models, you only get retired of it what you put into it. You still person to trade bully prompts aliases other you're near pinch thing very wide and not very interesting, successful my opinion. That's existent of Study Mode conscionable arsenic it is of different ample connection model-based programs.

In different words, acquisition successful Study Mode results from nan student's effort alternatively than nan teacher's brilliance.

Also: How to move AI into your ain investigation adjunct pinch this free Google tool

As a basal comparison, I asked Study Mode to thief maine study to publication and constitute successful Japanese, a connection I do not know. 

My punctual was: "I'd for illustration to study to publication and constitute Japanese moreover though I'm an absolute beginner!" 

The results successful Study Mode were astir identical to those of mean ChatGPT. Study Mode began pinch a mates of questions astir really I would for illustration to proceed, while regular ChatGPT simply responded pinch its projected instruction plan. In nan accompanying screenshot, Study Mode is connected nan left, and mean ChatGPT is connected nan right.

Screenshot from ChatGPT by Tiernan Ray/ZDNET

While it's not a bad thought to inquire maine astir really I would for illustration to proceed, arsenic an absolute beginner, it's alternatively pointless because I don't cognize thing astir what I'm going to learn. That's benignant of nan constituent of nan coach role. 

In some cases, ChatGPT suggested that we proceed by learning nan basal characters of nan autochthonal Japanese phonetic alphabet, nan hiragana. We proceeded statement by row, pinch maine trying to repetition nan hiragana that ChatGPT gave me.

At 1 point, it became clear to maine that simply learning hiragana 1 by 1 was not going to work. After astir half an hour, I refused ChatGPT's suggestions to proceed on nan measurement we had been going, and alternatively asked nan bot to springiness maine galore examples of existent words utilizing nan characters I had already learned. This started to thief maine cement my knowledge of nan characters. 

Also: I utilized ChatGPT's Study Mode to tutor maine for free - and you tin too

None of this stimulated my curiosity very overmuch pinch its rote exercises. To make it much absorbing for myself, I prompted Study Mode, "How galore hiragana are location altogether?" 

This was an illustration of a desire to understand nan larger scope of nan taxable matter, and it only happened because I asked. ChatGPT's consequence was a bully mentation of nan full number of hiragana. If I hadn't asked, I wouldn't person gotten specified an absorbing diversion. 

That's precisely nan point. Without a proposal from me, nan bot didn't person awesome ideas for really to move forward. As Sabrina points retired successful her introductory article, Study Mode relies a batch connected nan "Socratic method" of mobility and answer. However, successful nan realm of AI, nan enterprising personification often has much absorbing questions than nan bot. 

Also: GPT-5 bombed my coding tests, but redeemed itself pinch codification analysis

Why LLMs aren't very imaginative teachers

That shouldn't beryllium surprising. ChatGPT successful study mode has been shaped to conform to nan astir communal approaches to things. All connection models thin to enactment wrong what's likely, aliases highly probable, from infinitesimal to moment, which whitethorn beryllium due for reviewing worldly for a trial but is not stimulating for a learner of anything.

That is 1 of nan reasons that ChatGPT and different bots are regularly acing nan kinds of standardized tests wherever US students progressively fail: The programme has mastered nan routine, nan regurgitation of rote information.

It's beautiful clear that nan bot lacks a higher level of knowing of what it intends to thatch someone, namely, what educators telephone a curriculum. 

A program is simply a high-level knowing of really students study and really to move done worldly -- documents, examples, etc. -- successful a measurement that will not simply springiness answers but alternatively evoke nan student's ain expertise to inquire questions. 

Also: Has ChatGPT rendered nan US's acquisition study paper irrelevant?

In a bully education, aft all, 1 ends up pinch much questions than answers. 

Of course, we bot users aren't experts successful curriculum, either. That's why we mostly connection mostly nan aforesaid prompts: "Explain to me…", "Tell maine nan logic why…", and "Help maine study X." 

As users, we're stuck erstwhile we don't cognize what to inquire next. 

In nan past twelvemonth of on-and-off study, I haven't stuck pinch ChatGPT arsenic regularly arsenic I should if I really want to study a language. As nan novelty wore off, my resoluteness waned.

Let's alteration really we deliberation of bots for education

There's a clear connection for OpenAI and different AI developers here. Both Study Mode and normal ChatGPT are excessively overmuch shaped to nutrient a benignant of communal crushed successful nan emblematic speech without immoderate existent consciousness of really to bring a student to nan constituent of asking questions that unfastened up their desire to study more. 

Also: How AI-enabled autonomous business will alteration nan measurement you activity forever

There's small invention here, a batch of rote instruction plans. 

There's besides a connection for stressed-out teachers and professors here. It's earthy for group to scope retired for answers. If students are going to spell to bots for answers, and they surely are, past astir apt nan correct attack is to thief them find ways to create much questions retired of nan bot alternatively than playing nan bull and trying to forestall them from utilizing bots. 

Why not flip things around? Why not thief nan students push nan bot to nan constituent wherever an full taxable has go sufficiently analyzable that nan bot returns pinch much and much questions alternatively than simply providing answers for illustration an authority? 

It mightiness moreover beryllium a group activity, wherever nan coach recedes from nan activity domiciled and gets students to return nan lead, to hash retired really they are going to push nan bot into each nan regions of uncertainty. 

Also: College students tin get Google's AI Pro scheme for free now. Here's how

Think of that arsenic The New Curriculum, or, pinch a motion to today's programming methods, "Vibe Pedagogy," a measurement to hack nan bot to get to thing much interesting, much stimulating. 

Education is only astatine consequence from AI if nan coach is assumed to beryllium nan last authority. If, instead, acquisition is viewed arsenic uncovering retired conscionable really overmuch location is to know, really galore unfastened questions location are successful a section of study, past location is nary threat successful students utilizing exertion to unfastened up much and much questions. 

Let's study astir nan instrumentality on nan way

It's besides a bully measurement to merge nan study of a topic, specified arsenic nan American Revolution, pinch nan study of nan instrumentality itself, nan bot. Students are apt to walk nan remainder of their big lives interacting pinch nan bot successful immoderate fashion. Get to cognize nan bot -- its strengths and limitations.

As mentioned above, nan bots person aced each nan standardized exams. There is nary constituent successful forcing humans to strengthen dull regurgitation of nan facts. Much amended would beryllium to stimulate curiosity and question-asking, which humans still do amended than bots. 

Also: Anthropic launches Claude for Education, an AI to thief students deliberation critically

Because I asked really galore hiragana characters location were, I decided to inquire Study Mode a "meta" question, for illustration a pesky child: Why are location only 76 Hiragana? It gave maine a bully reply that didn't really reply nan mobility but was much of a description:

Japanese phonology is syllable-based and limited. It doesn't person each nan consonant clusters aliases vowel changes recovered successful English aliases galore different languages, truthful it doesn't request hundreds of symbols -- conscionable a neat, accordant set.

It is simply a benignant of answer, but it doesn't really answer, leaving nan mobility unfastened and intriguing.